Chapter 12: Path of the Artificer

The third-floor landing of Building 7 smelled like rust, iodine, and the particular brand of exhaustion that came from seventy-two hours without real sleep. Zhang Wei sat with his back against the stairwell railing, knees drawn up, the class selection prompt hovering at the edge of his vision like a pop-up ad he couldn't close.

Below, on the second floor, Fang Lei's people talked in low voices. Above, on the fourth, Chen Gang's group had gone quiet — Liu Peng's breathing had finally evened out, the re-splinted arm propped on a rolled-up jacket. Somewhere in between, Lin Mei was probably still awake. She slept like a cat.

Sixty-eight hours, forty-one minutes. The timer counted down in clean blue digits. After that, the System would assign him a class at random, and random in software meant worst-case scenario by default.

He needed to choose.

Zhang Wei had been stalling. Not because he lacked data — the terminal from the commercial district had burned the full class registry into his memory, and Code Sight's meta-tags from yesterday were layered on top, highlighted across every entry. He had enough data. The problem was that every time he opened the class selection screen, something about it made him want to throw up.

Not the stakes. The naming.

He pulled up the class menu.

Twelve options materialized in his peripheral vision, arranged in a grid that would have been orderly if the names made any kind of sense. They didn't. They were a catastrophe.

Warrior

Sword Sovereign of the Ninth Heaven

Ranger

Techno-Magus Enginseer

Healer

Battle Commander Lord

Shadow Blade

Spirit Caller of the Verdant Wilds

Oracle of the Endless Gaze

Dimensional Rift Walker Supreme

Null (Nullifier the System Breaker)

Grimoire Artificer

Zhang Wei read them twice. Then a third time, because he was certain he'd hallucinated one.

He hadn't.

"Warrior." He said it out loud, testing the syllables. Next to "Sword Sovereign of the Ninth Heaven," it looked like someone had given up halfway through the naming process. One was a placeholder from a 1990s CRPG. The other was Chinese cultivation webnovel bombast, complete with the "Heaven" suffix every xianxia protagonist accumulated the way frequent flyers collect miles.

He scrolled down. "Healer" sat directly below "Techno-Magus Enginseer." One was so generic it could have come from a Flash game. The other was straight out of Warhammer — the kind of name that only made sense if you'd spent three hundred hours painting miniatures and memorizing lore wikis.

"Shadow Blade" was clean. Japanese isekai energy. Short, evocative, marketable.

"Spirit Caller of the Verdant Wilds" was a JRPG boss class with a name longer than most countries' visa application forms.

"Oracle of the Endless Gaze" reeked of mobile gacha — the sort of character that cost $99.99 per pull.

"Dimensional Rift Walker Supreme" — Korean solo leveling all the way. The word "Supreme" added exactly nothing except syllable count.

And "Null (Nullifier the System Breaker)" couldn't decide if it was one name or two, so it included both, parenthetical and all — the food-label equivalent of listing every allergen in case of lawsuits.

"Battle Commander Lord" was three titles stapled together. One would have been enough.

Zhang Wei tried to query the System. Focused on the menu, pushed a question through the interface: Why are the class names inconsistent?

FEATURE_NOT_AVAILABLE_IN_CURRENT_BUILD. CLASS_NAME_DISPLAY IS VERSION_LOCKED.

"Version locked," he repeated. "The class names are version locked."

He tried again. Can I see the original developer names?

QUERY REJECTED. THIS FEATURE REQUIRES ADMINISTRATOR ACCESS.

"I don't want administrator access. I want the names to stop looking like they were translated by twelve different bots."

THANK YOU FOR YOUR FEEDBACK. YOUR INPUT HAS BEEN RECORDED.

The notification vanished. No follow-up. No ticket number. No confirmation beyond the System equivalent of a rubber stamp.

Zhang Wei pressed his palms against his temples. Thirty years of game development across four studios, and he had never seen shipping code with this level of localization inconsistency. This wasn't a rough translation. This wasn't even a bad translation. This was what happened when twelve separate AI models each translated a class name from an unknown source language into English without cross-referencing each other or any style guide.

He opened his eyes and activated Code Sight.

The class grid overlaid with translucent developer data — and beneath each chaotic player-facing name, Code Sight revealed the actual class IDs. Clean. Internal. Consistent.

1. Warrior → [CLASS_ID: Vanguard]
2. Sword Sovereign of the Ninth Heaven → [CLASS_ID: Striker]
3. Ranger → [CLASS_ID: Ranger]
4. Techno-Magus Enginseer → [CLASS_ID: Mechanist]
5. Healer → [CLASS_ID: Mender]
6. Battle Commander Lord → [CLASS_ID: Marshal]
7. Shadow Blade → [CLASS_ID: Shade]
8. Spirit Caller of the Verdant Wilds → [CLASS_ID: Warden]
9. Oracle of the Endless Gaze → [CLASS_ID: Oracle]
10. Dimensional Rift Walker Supreme → [CLASS_ID: Riftwalker]
11. Null (Nullifier the System Breaker) → [CLASS_ID: Nullifier]
12. Grimoire Artificer → [CLASS_ID: Arcanist]

"There they are." The real names. The names someone — some thing — had used when building the skill trees, writing the balance equations, designing encounter logic. The names behind the curtain.

Someone had built twelve internally consistent classes with clean developer IDs. Then a completely different process had generated the player-facing names with zero coordination. The backend was professional. The frontend was an afterthought. Or the frontend had been produced by a separate subsystem entirely — one that sampled human gaming data from multiple cultures and attempted to generate "culturally resonant" display names.

Badly.

Zhang Wei filed that away. Backend/frontend split meant the System's internal architecture and its player-facing interface were loosely coupled. The class names were garbage, but the class data underneath them might be solid.

Or the garbage ran deeper than names.

He started with what the System wanted him to pick.

Code Sight's meta-tags were still cached from the terminal, but now they came enriched with developer comments, version markers, and performance statistics that the terminal had kept behind another layer of encryption. Dense data. Zhang Wei parsed it like he was reviewing a pull request at four in the morning before a major release.

Vanguard [META_VIABLE] // Pick rate: 18.7% — within expected range // Survival rate at Lv30: 73% — highest of all classes // FIXME: damage scaling plateaus at Lv50-60; needs rebalance pass // TODO: add visual feedback for Iron Commander taunt radius

Solid class. High survival. The System wanted players picking Vanguard because a tank that held aggro and absorbed damage was predictable. Designing encounters around predictable classes was easy. That was the entire point.

Striker [META_VIABLE] // Pick rate: 22.1% — most popular class, expected // DPS output: highest raw single-target in game // WARNING: Berserker specialization causes rage-loop exploit at Lv35+ // DEPRECATED_SPECIALIZATION: Berserker removed in v0.3.2

Berserker had been deprecated inside the class. Not the whole class — just one branch. The rage-loop exploit sounded like the kind of thing someone had patched hastily and hoped nobody would notice before the next version.

He skimmed faster. Ranger — functional, no flags. Mender — high demand, Lin Mei's territory. Shade — "high skill ceiling, low floor — expect player frustration."

Oracle [META_VIABLE] // Pick rate: 3.1% — low but expected for support archetype // Specialization: Diviner — visions correlate with upcoming patch data at 67% accuracy // NOTE: Diviner visions cause neurological stress in host; scaling unresolved // TODO: reduce vision-induced migraine severity (BLOCKED — requires neuro compatibility data)

Huang Zixin's class. The visions that gave her migraines and nosebleeds weren't a design flaw — they were an unresolved scaling issue. The System knew its Oracle class hurt its users and had marked the fix as blocked.

Then he reached the DEPRECATED classes.

Mechanist [DEPRECATED] // DEPRECATED in v0.3.2 — causes skill tree corruption when construct AI interacts with patch-level environment updates // Pick rate: 2.1% — dropping // Survival rate at Lv30: 31% — unacceptable // NOTE: Do NOT buff this class without full audit of construct pathfinding // TODO: schedule removal from class pool in v0.5 (BLOCKED — requires Architect approval) // CROSS_REF: see [Arcanist/Artificer] for orphaned cross-module dependencies

Zhang Wei read it twice. Mechanist constructs broke when patches updated the environment — the AI pathfinding couldn't adapt to new terrain layouts. Every monthly patch was a potential wipe. And someone had flagged the class for removal but couldn't get it approved.

The System's internal bureaucracy had cracks.

He considered Mechanist for exactly four seconds. Unpatched blind spots sounded appealing, but the construct problem was fundamental. Not an exploit he could leverage — a broken feature he'd be building on top of. Like constructing a house on a foundation that actively collapsed during rainstorms.

Not an opportunity. A trap.

But the last line snagged him. CROSS_REF: see [Arcanist/Artificer] for orphaned cross-module dependencies. Mechanist had orphaned dependencies inside Arcanist's tree? He filed it for later and moved on.

Warden [DEPRECATED] // DEPRECATED in v0.3.1 — nature mana pool incompatible with urban terrain generation // Pick rate: 1.4% — lowest of all classes // TODO: implement fallback mana source for non-natural environments // Balance team has marked as LOW_PRIORITY

Dead in the city. Nature magic needed natural terrain, and the System had replaced most of it with corrupted architecture. Forty percent mana regeneration penalty for urban Warden players. The balance team had marked the fix as low priority, which in development meant it would never be fixed.

DEPRECATED wasn't freedom. It was neglect that happened to hurt players.

He moved on to the EXPERIMENTAL classes.

Marshal [EXPERIMENTAL] // Pick rate: 0.8% — insufficient data // TODO: finalize formation buff stacking rules

Riftwalker [EXPERIMENTAL] // Pick rate: 1.1% // FIXME: teleportation causes zone-loading failures in dense urban areas // NOTE: implement fail-safe for failed teleport (currently: instant death)

Instant death on a failed teleport. "Currently: instant death." Not flagged as critical. Not marked urgent. Just a note, sitting in the comments. As if player death from a bugged skill was an acceptable known issue to be addressed in a future sprint.

Nullifier [EXPERIMENTAL] // Pick rate: 0.3% — rarest class // Ability_Copy skill tree: 40% complete // TODO: add remaining skill branches (BLOCKED — pending playtest data)

Forty percent complete. Half-built. The class literally didn't have all its abilities implemented.

Zhang Wei pushed a question at the System: Can I preview the full Nullifier skill tree?

SKILL_TREE_PREVIEW: FEATURE NOT AVAILABLE IN CURRENT BUILD. REASON: SKILL DATA EXCEEDS TIER_1 ACCESS CLEARANCE.

"Of course it does." He tried a different angle. What about Arcanist? Can I see the full specialization tree?

ARCANIST SPECIALIZATION TREE: AVAILABLE. SPECIALIZATIONS: [REDACTED UNTIL CLASS SELECTION]

I need to see the specializations to make an informed choice.

CLASS SELECTION IS INTENDED TO BE MADE BASED ON BASE_CLASS DESCRIPTION. SPECIALIZATION DETAILS ARE UNLOCKED AT LEVEL 40.

You can't evaluate a class without understanding its endgame path. This is a fundamental design flaw.

CLASS SELECTION IS NOT A DESIGN EVALUATION. PLEASE SELECT YOUR CLASS.

The System's chatbot response carried the cheerful dismissal of customer-service AI trained on "How can I help you today?" and nothing else. The dialogue tree was shallow — two branches deep, then a hard stop. No contextual follow-up. No ability to drill down. The interface was designed for players who accepted what they were given.

Zhang Wei let it go. The terminal data was cached, and Code Sight could fill the gaps.

He closed his eyes and replayed the Arcanist tree from memory. Three specializations at Level 40: Elementalist, Spellweaver, and a third path that Code Sight's metadata had flagged with an unusual comment buried in the base class entry.

Arcanist [META_VIABLE] // Pick rate: 14.2% — expected for caster archetype // Specialization paths: 3 (2 active, 1 pending review) // TODO: resolve Artificer_path cross-module dependencies // NOTE: mana framework stable; spell scaling within balance targets

"Two active, one pending review." The third path was stuck in development limbo. Not deprecated, not experimental — pending. Like a feature that had been submitted but never merged.

He focused on that third path and pushed Code Sight deeper.

ARTIFICER [SPECIALIZATION: Arcanist > Artificer] // Status: ABANDONED // Last updated: v0.1 // Pick rate: 0.00% — no player has selected this specialization // Developer: [REDACTED] // NOTE: Cross-port from [Mechanist/Construct_Analysis] module during alpha integration // NOTE: Skill tree references deprecated [Mechanist/API_v1] endpoints // NOTE: Arcanist mana framework partially incompatible with construct-spawn subroutines // TODO: complete integration OR revert to Mechanist tree // Status note: BLOCKED since v0.1 — pending Architect review // Priority: lowest

Zhang Wei stared at the comments.

Not deprecated. Not experimental. Not META_VIABLE.

Abandoned.

Artificer had been built in version 0.1 — the first version of the System, the skeleton that everything else had been layered on top of. It was a cross-port from Mechanist's Construct_Analysis module, slotted into Arcanist's specialization tree during alpha integration. Someone had taken Mechanist's component analysis code, grafted it onto Arcanist's mana framework, and left it there when the System shipped to reality.

It had never been updated. Never been balanced. Never been touched by a single patch since the day it was committed.

No player had selected it, because the System hadn't promoted it. The skill tree was marked incomplete. It sat in the backlog — dead code that nobody remembered writing.

And here was the layer that made his breath catch.

The System's monitoring algorithms tracked META_VIABLE classes because those were popular. They tracked DEPRECATED classes because those needed removal. They tracked EXPERIMENTAL classes because those were unstable.

Artificer was none of those. It was miscategorized — classified as an Arcanist specialization but built from Mechanist's design tree. The System's monitoring scanned Arcanist for Arcanist-pattern behavior. Artificer didn't match Arcanist patterns because it was actually Mechanist code. The monitoring algorithms wouldn't flag it because they were searching for known patterns, and this pattern had never been seen before.

Two layers of neglect. Not just forgotten — misplaced. A Mechanist component living inside an Arcanist's tree, invisible to both classes' monitoring protocols. Like finding a backdoor that was left in a production build because it was accidentally committed to the wrong branch and never got merged into the security review.

In twenty years of game development, Zhang Wei had learned one principle that trumped all others: the most dangerous code in any system was the code nobody remembered. Not the deprecated features — those were tracked, documented, eventually purged. Not the experimental builds — those were monitored, evaluated, subject to review cycles. The dangerous code was the stuff from the original commit that sat in production because nobody had filed a ticket to remove it.

Nobody monitored forgotten code. Nobody wrote anti-exploit measures for forgotten code. Nobody balanced patches against forgotten code.

And this code wasn't just forgotten. It was in the wrong directory entirely.

For someone trying to operate beneath the System's radar, Artificer wasn't just a good choice. It was the only choice that made sense.

Not because it was powerful — it probably wasn't. Not because it synergized with Code Sight — he hadn't tested that yet. But because choosing Artificer meant choosing a path the System wasn't watching. A blind spot nested inside the System's own awareness of its players. A Mechanist-derived capability hidden inside an Arcanist class, accessible through Arcanist's mana system without triggering Mechanist's deprecation flags.

He dug deeper. Code Sight pushed past the surface metadata into the Artificer skill tree itself.

ARTIFICER SKILL TREE (PARTIAL): Lv10: Disassemble Lv15: Component Analysis Lv20: [DATA NOT FOUND] Lv25: [DATA NOT FOUND]

Two skills out of an undefined tree. The rest was missing — never implemented, never data-filled. But two skills were enough to start.

Then Code Sight found something else. A branch within Artificer that wasn't listed on any preview. Hidden behind the incomplete tree, visible only through Code Sight's layered analysis:

HIDDEN_SKILL_TREE_DETECTED: "System Architecture Reverse Engineering" STATUS: [REQUIRES ARTIFICER SELECTION TO UNLOCK] NOTE: Ported from [Mechanist/Reverse_Engineering] module during alpha WARNING: Interfaces with deprecated API endpoints — access may trigger monitoring anomalies

Zhang Wei read the warning twice. Interfaces with deprecated API endpoints. Access may trigger monitoring anomalies.

The System knew this skill tree existed. It had flagged a warning. But the warning was buried inside an abandoned specialization inside a META_VIABLE class — three layers deep in a hierarchy that nobody was watching because nobody had selected the path.

Forgotten code. Miscategorized code. And buried inside that, a hidden skill tree that connected directly to the System's deprecated API layer.

For a game developer who had spent his career understanding systems from the inside, this was the equivalent of finding an unlocked door to the server room in the basement of a building everyone thought was empty.

Zhang Wei opened his eyes. The class selection grid still floated in front of him, twelve names ranged across his vision. Most of them were still terrible.

He focused on "Grimoire Artificer." Code Sight overlaid the clean ID: Arcanist. Beneath it, the specialization path he'd committed to memory: Artificer. Abandoned since v0.1. Cross-pollinated from Mechanist. Zero pick rate. Zero monitoring.

The System had named its Arcanist display version "Grimoire Artificer" — artifice, craft, the act of making. But the actual Artificer specialization within that class had been abandoned. The name promised creation. The code never delivered it.

He was going to deliver it himself.

Let me see the Disassemble skill details, he thought, pushing at the System.

For a long moment, nothing happened. Then:

DISASSEMBLE SKILL DATA: DESCRIPTION: Deconstruct System-spawned objects into component layers. Reveals material composition, modification protocols, and interface architecture. REQUIREMENTS: Physical contact with target object. Mana cost: 6 per use. RANGE: Touch NOTES: Non-destructive analysis. Object can be reassembled after decomposition.

Physical contact. Six mana per use. Non-destructive. He'd need to touch something System-spawned, spend six mana, and the object would split into visible layers he could analyze. His mana pool was twenty-four. Four uses before he ran dry.

He thought about the selection prompt. Twelve options. The System wanted him to pick Vanguard, Striker, one of the META_VIABLE classes where the balance team could track his progress. Where the monitoring algorithms expected players to go.

He thought about the Artificer tree. Two skills. A hidden branch. Deprecated API access. Zero monitoring. Zero pick rate. Zero balance adjustments.

No, he thought. I'll take the incomplete tree.

ARE YOU SURE? ARTIFICER IS AN UNBALANCED, UNTESTED SPECIALIZATION PATH WITH INCOMPLETE SKILL DATA.

"Perfect," Zhang Wei said. "That's exactly what I want."

He exhaled slowly. The stairwell was quiet. Fang Lei's people had gone to sleep, and the only sound was the distant hum of System corruption pulsing through the building's walls.

He focused on the selection prompt and made his choice.

CLASS: ARCANIST [CONFIRMED] SPECIALIZATION: ARTIFICER (PENDING — FULL TREE UNLOCKS AT LV40 — BASE SKILLS AVAILABLE)

The confirmation didn't come as a pop-up. It came as a vibration that started in his fingertips.

Not his spine. His hands.

A hum, bright and immediate, the sort of sensation that travels through skin into bone. It crawled up his fingers into his palms, pooled in his wrists, then shot up his forearms in branching paths that felt like wiring being installed in real-time. His fingers flexed involuntarily. Every joint popped — not painfully, but with the crisp precision of a mechanism clicking into alignment.

His hands felt longer. More articulated. As if new tendons had been threaded through them. The urge to touch something was overwhelming — to feel its surfaces, find its seams, discover how it came apart. His fingers itched with a hunger that had nothing to do with pain and everything to do with purpose.

His vision changed next. Not a gradual shift — a hard refresh, like a display switching resolutions. The stairwell snapped into new clarity, but it wasn't sharper. It was deeper. He could see layers. The concrete wall in front of him wasn't just a surface — it was a composite. Aggregate, binding agent, reinforcement mesh, paint layer. He could feel how they connected, how they were assembled, what tools would separate them.

Not see. Feel. The knowledge lived in his hands, not his eyes. His fingertips registered stress points, material boundaries, structural weaknesses — information delivered through touch rather than sight, the way a master carpenter knows a joint is loose by tapping it.

Blueprints flickered through his awareness. Not images — intuitions. He understood, without knowing how he understood, that everything System-touched had an architecture. A build order. A design specification. And his hands — his newly rewired, humming, electric hands — knew how to read that architecture through contact.

The heat settled in his palms. A low, steady warmth, the way a soldering iron gets after hours of continuous use — hold it carefully or burn. His mana pool materialized in his awareness — not behind his sternum like a traditional caster reservoir, but distributed through his hands and forearms, a network of energy nodes clustered around his fingertips and palms. Twenty-four points. Low. The starter pool for a Level 10 Arcanist with no equipment bonuses.

The transformation lasted eight seconds. It felt like being assembled.

When it ended, Zhang Wei looked at his hands. They were the same hands. Same scars, same bitten nails, same calluses from a decade of keyboard work. But they felt different. Capable. Hungry.

The world looked different too. No syntax lines. No luminous grammar tracing every surface. Instead, he saw the world as a diagram of itself — components and connections, interfaces and protocols. The stairwell railing wasn't just metal; it was a load-bearing structure with stress points, material composition, and System-modified molecular bonds. The concrete floor was an assembly of aggregate and binder with embedded corruption nodes where the System had rewritten its molecular structure.

Code Sight still worked — the statistical overlay was intact. HP bars, damage values, hidden tags. But layered beneath that, running through his enhanced perception like a second operating system, was the architectural view. Protocols. APIs. Hardware interfaces. The System wasn't written on reality in syntax anymore. It was built into reality as infrastructure.

CLASS SELECTED: ARCANIST SPECIALIZATION: ARTIFICER (PENDING — FULL TREE UNLOCKS AT LV40) BASE STATS MODIFIED: +4 INT, +2 WIS, +2 PER, +1 VIT NEW SKILL: Disassemble Lv.1 NEW SKILL: Mana Sense Lv.1 UNIQUE ABILITY UPDATE: Code Sight — System Architecture Overlay ENABLED

Footsteps on the stairs. Chen Gang descended two steps and stopped, reading Zhang Wei's posture the way he'd read terrain for thirty years.

"Your hands are doing something different," Chen Gang said.

Zhang Wei looked down. His fingers were moving — flexing, tapping, pressing against his own palm in a pattern that felt like testing a mechanism. He hadn't told them to do that.

"I feel different," he said. "Picked a class."

"Which one?"

"Artificer."

Chen Gang waited for more. Zhang Wei didn't provide it.

"Get some sleep," Chen Gang said. "Whatever you just did, you look like you need eight hours."

Chen Gang climbed back upstairs. Zhang Wei heard him settle into his watch position on the fourth-floor landing — the same spot every night since the Stress Test. Old habits from a career where sleep was optional and vigilance wasn't.

Zhang Wei gave him five minutes. Then he descended to the ground floor, slipped through the barricade, and stepped outside.

Midnight. The street was dark — System corruption had killed the streetlights on Day 1. Moonlight filtered through the clouded sky, painting everything in shades of grey. The architectural overlay traced every surface in faint amber lines — not code to be written, but structure to be understood. A world built from components, connected by protocols, running on infrastructure he could now feel beneath his fingertips.

He sat on the curb outside Building 7 and unzipped his pack. Inside, among the salvaged supplies, was the Riftclaw carapace fragment — a piece of armored hide the size of his palm, cracked and dull, dropped when the creature died yesterday. Chen Gang had told everyone to collect monster drops. SP material, he'd called it. Zhang Wei had shoved it in his pack and forgotten about it.

He held it in both hands.

His fingers knew what to do before his brain caught up.

They found the seams — hairline fractures in the carapace where the material composition shifted from organic chitin to System-modified composite. His thumbs pressed instinctively at stress points. He felt the fragment resist, then yield, the clean give of a sealed case popping along its designed opening.

Disassemble.

The word appeared in his awareness — not as a notification, but as a description of what was happening. His hands pulled the fragment apart along seams he could feel but not see. Three layers separated cleanly: the outer carapace, a middle layer of crystallized energy substrate, and a core of dense material that pulsed with residual System data.

He held the three layers apart and looked.

Without Code Sight, the layers were just material — chitin, crystal, dense core. Interesting, but unremarkable.

He activated Code Sight.

The world doubled. And this time, the doubling was different from anything he'd seen before.

Code Sight had always shown him the statistical overlay — HP bars, damage values, hidden tags. Now it showed those AND the structural architecture beneath them. The three layers of the Riftclaw fragment lit up with data he'd never seen on any System object:

Outer layer:

MATERIAL: Organic chitin (SYSTEM_MODIFIED) MODIFICATION: Hardening protocol v2.1 — molecular restructuring for armor function FUNCTION: monster_part.convert_to(SP) on decomposition INTERFACE: StandardMonsterComponent_v2

Middle layer:

MATERIAL: Crystallized mana substrate FUNCTION: Energy storage — residual power from Riftclaw ability pool INTERFACE: ManaReservoir_v1 (DEPRECATED — endpoint still functional)

Core:

MATERIAL: Unknown composite (SYSTEM_NATIVE) FUNCTION: Identification tag — links to spawn_table entry Riftclaw_08 INTERFACE: SpawnTracker_v2

Zhang Wei's breath caught.

He was looking at the System's component architecture. Not stats. Not numbers. The actual design specification for how this object was built. Material layers, function calls, interface endpoints — the same abstraction layers he'd spent twenty years working with in game engines.

And the interfaces were standardized. StandardMonsterComponent_v2. ManaReservoir_v1. SpawnTracker_v2. Versioned APIs. Modular design. The System built its monsters from plug-and-play components, each with a defined interface, each upgradeable independently.

This was engineering. Deliberate, structured, systematic engineering. Someone had designed this architecture with the same principles he'd used building game systems — modular components, standardized interfaces, versioned APIs.

His Artificer intuition filled in the gaps Code Sight couldn't see. The connections between layers — how the hardening protocol interfaced with the mana substrate, how the spawn tracker communicated with the System's central database, where the stress points were that would cause the whole assembly to fail if disrupted. He couldn't read this information. He could feel it, the way a mechanic feels where a bolt will strip or a weld will crack.

Together, Code Sight and Artificer gave him a complete picture. Code Sight showed what the object was made of. Artificer showed how it was built. Combined, they revealed why — the design intent, the engineering choices, the systematic logic behind every component.

Two abilities that had never been tested together were producing emergent perception that the System's balance models didn't account for. The synergy wasn't a damage modifier or a combat exploit. It was understanding. Complete, structural, architectural understanding of how the System built its reality.

He pressed the layers back together. They reassembled with the clean click of a case sealing shut. The fragment was whole again. Undamaged. The Disassemble skill was non-destructive — analysis, not demolition.

He sat on the curb, holding the reassembled Riftclaw fragment, and stared at his hands.

The mana cost had been six points per use. He'd spent six on the disassembly, and the Code Sight integration during analysis had drained another three from passive regeneration. Eleven points consumed. Thirteen remaining. Enough for two more disassemblies if he was careful.

Not a combat build. Not a damage build. An information build. The power wasn't in what he could destroy — it was in what he could understand.

And understanding, in a world run by rules that most people couldn't see, was the most dangerous power there was.

He tucked the fragment back into his pack and stood. The moon had shifted. He'd been sitting there longer than he'd intended.

He turned toward Building 7.

And saw the red notification.

Not blue. Red. The color the System used for alerts that weren't meant for players.

SYSTEM_WATCHDOG — ALERT LEVEL: ESCALATED SUBJECT: ZHANG WEI [PLAYER_ID: 000000000001] FLAG: [ANOMALY] TRIGGER: CODE_SIGHT + ARTIFICER SYNERGY — UNINDEXED INTERACTION DATA EXPOSURE: System component architecture accessed via deprecated API endpoint MONITORING PROTOCOL: ACTIVE INTERVENTION THRESHOLD: 3/10 CURRENT ASSESSMENT: OBSERVATION

Zhang Wei read it twice.

The trigger was specific: not just the synergy itself, but the data it exposed. He'd accessed System component architecture through a deprecated API endpoint — the ManaReservoir_v1 interface on the Riftclaw fragment. That endpoint was deprecated but still functional. The System tracked access to deprecated endpoints because they were potential vulnerability surfaces.

He hadn't been flagged for the Artificer ability itself. He'd been flagged for what the ability let him see.

Different problem. Worse problem.

The System could tolerate players using unexpected abilities. It could not tolerate players reading its component architecture. That was internal documentation. Build specifications. The kind of data that, in a game, would only be accessible to developers.

He was reverse-engineering the System's source code. And the Watchdog had noticed.

Intervention threshold: three out of ten. Current assessment: observation. He had room. Some room. But the Watchdog was monitoring the specific pattern now — Code Sight plus Artificer, deprecated API access, component architecture exposure. Every time he used the combination, the threshold would tick higher.

Zhang Wei dismissed the red notification. Closed his eyes. Let the mana regenerate one point at a time.

He climbed back through the barricade. Up the stairs. Settled onto the third-floor landing.

Outside, something howled — distant, high-pitched, Level 6 or higher. The corruption nodes on the building's walls pulsed in response, adjusting, recalculating their load distribution. The world shifted under the System's attention, and somewhere in the background architecture, a monitoring algorithm updated its priority queue.

Zhang Wei was no longer a background process.